test: harden TestE2E_PushDLQLifecycle against CI load

[7schmiede]

Problem

TestE2E_PushDLQLifecycle intermittently fails on loaded CI runners (ubuntu-latest) with:

requeued message not re-delivered within 10s (deliveryCount=3, before=3)

Root cause is a TOCTOU race in the test harness, not a too-short deadline. The re-delivery check snapshotted beforeRequeueDeliveries := deliveryCount.Load() after pd2.Start(), so under load pd2 could re-deliver the requeued message (n=3 → success → count becomes 3) before the snapshot was taken. The busy-poll then waited forever for count > 3, which explains the before=3, deliveryCount=3 timeout. Simply raising the deadline would not have fixed this.

Change

• Replace the racing sleep-poll with a deterministic signal: the target handler does a non-blocking send on a buffered channel on the successful (n>=3) delivery — which is exactly the re-delivery of the requeued message. The test waits on that channel with a 30s safety bound instead of racing a fixed wall-clock budget.
• Remove the racy beforeRequeueDeliveries snapshot.
• Bump the downstream queue-empty deadline 5s → 15s so the flake isn't relocated one assertion down.

The assertion stays meaningful — the requeued message must be re-delivered, or the test fails.

Scope

Test-harness robustness only. No dispatcher behavior change.

Verification

• go vet ./internal/e2e/ — clean
• go test ./internal/e2e/ -run TestE2E_PushDLQLifecycle -count=5 — passes all 5

🤖 Generated with Claude Code